CyberCertPrepAI Security FundamentalsSecure AI Deployment & MLOpsEASYFree question
Which inference deployment model keeps both the model weights and all prompt data entirely within an organization's own infrastructure?
- A.Calling a commercial SaaS LLM API over the internet
- B.Serverless GPU functions on a shared public cloud platform
- C.Self-hosting an open-weights model on servers the organization controls
- D.Embedding a third-party chatbot widget into the intranet
Why C is correct
Self-hosting open-weights models (for example with vLLM or TGI on company-controlled servers) is the only listed option where weights, prompts, and outputs never leave infrastructure the organization administers. SaaS APIs and embedded third-party widgets send prompt data to an external provider, and serverless GPU platforms run on provider-managed, typically multi-tenant hardware. This is a foundational tradeoff: self-hosting maximizes control at the cost of operational burden.
Know someone studying for AI Security Fundamentals? Send them this one.